How Aster's code is written and reviewed, where the earlier Aster code comes from, and what we can and cannot say about copyright in AI-assisted code.
How code is written
- Most of Aster's code is written by AI coding agents (large language models run as automated coding sessions) under the direction of the company's human developer, who sets the goals, specifications and acceptance tests. The git history records the agent identity in the commit author; agent commits are visible as such.
- Agents work on review branches.
mainis protected by a server-side hook: nobody, human or agent, pushes to it directly. Changes reachmainthrough a merge decided by the human maintainer or the integration process they run. - Changes are checked by automated gates (formatting, lints with warnings denied, tests) and, for significant changes, reviewed by a second agent or a human before merge. See
CONTRIBUTING.mdfor the rules. - Tests are not weakened or skipped to pass a change; this is a rule, and is enforced by review rather than by a tool.
Human-review policy
| Change type | Required review |
|---|---|
Anything touching security properties (the I/O guard, parsers of untrusted input, unsafe, dependencies, release and signing scripts) | Human maintainer reviews and approves before merge |
| Rendering or API behavior | Automated gates plus an independent review pass (agent or human); a human decides the merge |
| Documentation | Human reads before the release it ships in |
| Releases | Human approves and signs |
At present the number of human reviewers is one person. This is a limit of a small team and is the main gap recorded in SSDF mapping and draft attestation.
Origin of the earlier Aster code
Aster imports earlier Aster code (its original sources are C#), "through 37d9876 with its Git history" (see the root README.md). The earlier sources, fixtures and vendored dependencies remain in the repository as reference material, and the Rust implementation was developed from it using the C# code as a behavioral oracle. The first commits are authored by the project owner; the authorship and license status of everything in the imported earlier code is for the owner to confirm and record here (open item). The repository's LICENSE file still carries the imported MIT text with the earlier project's copyright notice. The SDK and examples are released under that MIT license; the runtime ships under the Aster Runtime EULA. Because the license names the earlier project's contributors, whether the notice should be updated, and whether all imported contributions can be licensed this way, is for the owner and counsel to confirm (open item).
Vendored third-party code lives under rust/vendor and is listed with its licenses in THIRD-PARTY-NOTICES.md and in the SBOM described in Supply chain: SBOM, licences, advisories, vetting, reproducibility.
Copyright status of AI-assisted code: a neutral note
Whether, and how far, copyright subsists in material produced with generative AI is unsettled and differs by country. The US Copyright Office's report Copyright and Artificial Intelligence, Part 2: Copyrightability (January 2025) concludes, in summary, that copyright protects human authorship; that prompts alone do not generally give a user enough control to be the author of the output; and that human contributions such as selection, arrangement, creative modification and human-written portions can be protected. It evaluates cases individually. (https://www.copyright.gov/ai/) Other jurisdictions have taken different or still-developing positions.
What this means for Aster, stated without legal conclusions:
- Some of Aster's code may have thinner copyright protection than conventionally written code, and the extent is not known to us.
- The MIT license grants permissions only to the extent the licensor holds rights. Where little or no copyright subsists in some code, that code may be freely usable by anyone regardless of the license; for users this does not reduce their rights, but it limits what we can enforce. The same limit applies to the Aster Runtime EULA: where little or no copyright subsists, we cannot enforce it. We rely on human selection, architecture, specification and modification of the code, and the recorded human direction in the history.
- We keep records that support a claim of human authorship: specifications, review decisions and the history.
- Users who need an opinion on this point for their own diligence should ask their counsel; we have not obtained a legal opinion. This is an open item.
Developer Certificate of Origin
Every commit by an authorized contributor carries a DCO 1.1 sign-off (git commit -s), as set out in CONTRIBUTING.md. For agent-written commits the sign-off is made by the human who directed the work. The sign-off is a recorded statement of origin, not a legal guarantee. Existing history predates this rule and is not retroactively signed.
Third-party and model-output risks we manage
- Dependencies and vendored code go through license review (Supply chain: SBOM, licences, advisories, vetting, reproducibility).
- We do not feed customer data to models; Aster has none (see No data: what Aster does and does not do).
- Model output that reproduces recognizable third-party code is treated as a defect, not accepted. No automated scanner for this is in place yet; that is a gap.